Future changes in precipitation extremes over Southeast Asia: insights from CMIP6 multi-model ensemble

Past assessments of coupled climate models have indicated that precipitation extremes are expected to intensify over Southeast Asia (SEA) under the global warming. Here, we use outputs from 15 climate models from the Coupled Model Intercomparison Project Phase 6 (CMIP6) to evaluate projected changes in precipitation extremes for SEA at the end of the 21st century. The results suggest that CMIP6 multi-model ensemble medians show better performances in characterizing precipitation extremes than individual models. Projected changes in precipitation extremes linked to rising greenhouse gas (GHG) emissions (represented by the latest proposed Shared Socioeconomic Pathways) increase significantly over the Indochina Peninsula and the Maritime Continent. Substantial changes in the number of very heavy precipitation days (R20mm) and the intensity of daily precipitation (SDII) indicate that such locally heavy rainfall is likely to occur over a short time and that more precipitation extremes over SEA are probable in a warmer future. This is consistent with projections from the Coordinated Regional Downscaling Experiment and CMIP5 models. The present study reveals the high sensitivity of the precipitation extremes over SEA, and highlights the importance of constrained anthropogenic GHG emissions in an ambitious mitigation scenario.


Introduction
How fast is too fast? The global mean temperature in 2019 was around 1.1 ± 0.1 • C above pre-industrial levels, which may have been the second warmest year on record (WMO 2019). Latest reports confirm that the global mean temperature is warming at a rate of 0.1 • C-0.3 • C per decade. Restricting global warming to below 2 • C (or the preferred lower limit of 1.5 • C) is the committed target of the United Nations Framework Convention on Climate Change (UNFCCC) Paris Agreement (UNFCCC 2015). To achieve this ambitious target under current anthropogenic greenhouse gas (GHG) emissions seems like a race against time. However, while the present fossil fuel investments and mitigation strategies do not support the strategy to keep global warming below 1.5 • C, the coronavirus disease 2019 (COVID-19) pandemic has dramatically led to a temporary reduction of both GHG emissions and air pollutants around the world since February 2020 (Smith et al 2019, Forster et al 2020, Le Quéré et al 2020. Despite this, looking ahead, the global cooling response to the pandemic is likely to be sudden and small, and would not drastically alter future increases in climate extremes. It has been well demonstrated that the current rapid global warming has tended to intensify precipitation extremes in tropical regions through changes in atmospheric water vapor content, circulation patterns, and moisture supply, which in turn significantly influence natural ecosystems, water management, and agriculture in less-developed countries (Allen and Ingram 2002, Zhai et al 2005, Held and Soden 2006, Vecchi et al 2006, Hulme 2016, Marotzke et al 2017, Sillmann et al 2017, Chen and Sun 2018, Nikulin et al 2018. Southeast Asia (SEA), an area characterized by a large coastal population in complex terrains, is generally considered to be one of the  et al 1987, Matsumoto 1992, Webster et al 1998, Lau et al 2000, Wang et al 2000, Zhu 2018). Simulating the past and present climate over SEA is therefore a scientific challenge. It is also crucial for providing information on future changes in precipitation extremes that can be used by local governments to implement adaptation and mitigation. Reliable projections are of great importance in projecting future climate change. The Coupled Model Intercomparison Project Phase 6 (CMIP6) uses the new scenarios, named Shared Socioeconomic Pathways (SSP), which are combined with the Representative Concentration Pathways (RCP) of CMIP5 (Eyring et al 2016). These new combinations enable several ways to examine the future projected changes in precipitation extremes over SEA. Current stateof-art climate models are more robust than previous CMIP ensembles and have shown the effective improvements in reproducing large-scale patterns of climate variables (Akinsanola et al 2020;Gusain et al 2020, Ha et al 2020, Jiang et al 2020, Wang et al 2020, Zhai et al 2020, Chen et al 2020b. However, to the best of our knowledge, there is still a lack of information regarding the projections of SEA precipitation extremes under the new CMIP6 scenarios. In this study, we aim to provide a comprehensive picture of the changing magnitude of precipitation extremes over SEA, and also to address the following questions: (a) How do the CMIP6 multi-models perform in simulating precipitation extremes over SEA? (b) What are the dominant roles of the projected precipitation extremes under CMIP6 scenarios in the long-term future?

Simulation data from the CMIP6 archive
The daily precipitation data are obtained from 15 CMIP6 model ensembles (table 1) for the first realization that are available at the time of initializing this study (up to August 2020). The future changes in precipitation extremes are projected under four SSP scenarios. For CMIP6, the four SSPs are categorized as SSP1-2.6 (sustainability), SSP2-4.5 (middle-ofthe-road), SSP3-7.0 (regional rivalry) and SSP5-8.5 (fossil-fueled development) (O'Neill et al 2016). The historical simulations for the reference period of 1985-2014 are used in this study, while a future period of 2071-2100 is selected for the analysis of climate projection. The model outputs are interpolated to a regular geographical grid of 1 • × 1 • , consistent with the observational precipitation dataset introduced in the following subsection.

Observational data and index representation of extremes
The daily gridded precipitation data from the Southeast Asian Climate Assessment and Dataset (SACA&D) covering the period of 1981-2017 are used in this study. This high-resolution dataset with a resolution of 1 • × 1 • , called SA-OBSv2.0, has undergone strict quality control procedures, including data homogenization and time consistency, to improve reliability. SA-OBSv2.0 dataset includes several meteorological variables, such as daily precipitation amount, daily mean temperature, daily maximum temperature and daily minimum temperature. The daily precipitation series are collected from the meteorological agencies of Australia, Indonesia, Malaysia, Philippines, Singapore, Thailand, and Vietnam. Additionally, precipitation series from the Global Historical Climate Network for American Samoa, Fiji, Kiribati, the Federated States of Micronesia, Papua New Guinea, Samoa, and the Solomon Islands are also involved (van den Besselaar et al 2017, Ge et al 2019a). As a result, 1393 rain gauge stations distributed over the whole SEA contribute to derive the SA-OBSv2.0 gridded dataset. We focus on the period of 1985-2014 to match the timescale of the CMIP6 historical runs. The SEA domain is located between 10 • S to 23 • N and 95 • E to 140 • E (figure 1) and comprises five subregions: the Indochina Peninsula Following the Expert Team on Climate Change Detection and Indices (ETCCDI, Zhang et al 2011, Sillmann et al 2013a, 2013b, six indices are selected to represent the precipitation extremes, as displayed in table 2. These are Rx1day (maximum consecutive 1 d precipitation), Rx5day (Maximum consecutive 5 d precipitation), SDII (Simple daily intensity), R20mm (very heavy precipitation days), CDD/CWD (Consecutive dry/wet days), PRCPTOT (Total precipitation of wet days) and R95pTOT (Precipitation of very wet days). More detailed information on the indices can be found on the ETCCDI website of http://etccdi.pacificclimate.org/indices.shtml.

Model performance metrics
The most common method to evaluate a climate model is the quantitative assessment of 'model-fit'; that is, how well the model results match observationbased data and results of other models or model versions. Here, we use the relative root mean squared error (RMSE ′ ) and signal to noise ratio (SNR) to quantify the empirical accuracy and the robustness of the ensemble results, which have been shown skillful in climate projections and multi-model assessment ( Given the substantial uncertainties of projected changes in precipitation, the RMSE ′ used to evaluate the climate simulation capability of each individual CMIP6 model is defined as follows (Gleckler et al 2008, Dong et al 2015. First, the RMSE is calculated for each model index with respect to the SA-OBS observations: With X being the model climatology of an extreme precipitation index and Y the corresponding index of the observation. All RMSEs are then used to derive the RMSE ′ of each model: To express the credibility of the projection ensemble results, the SNR is defined as follows (Zhu et al 2020a): where x i represents the variable or index simulated by an individual model, x e denotes the corresponding ensemble result, and n is the ensemble size; the numerator and denominator refer to signal and noise, respectively. Therefore, SNR > 1 implies that signal is greater than noise, indicating relatively reliable projections.

CMIP6 model validation
The The RMSE ′ of the ensemble medians are shown in the last column of figure 2. The model ensemble median is chosen to represent the deterministic ensemble result rather than the mean value, to avoid results influenced by abnormally large model errors (outliers). It is demonstrated that the ensemble median results perform better than any individual models for all indices, which eliminates the structural model uncertainties to a great extent and so can be considered to reasonably represent the future projections in the study. Figure 3 presents the projected changes of CMIP6 ensemble medians of precipitation-based indices over SEA during the long-term future at the end of the 21st century (2071-2100). Most of the extreme precipitation indices (i.e. Rx1day, Rx5day, SDII, R20mm and R95pTOT) increase significantly across the ICP and Maritime Continent for all scenarios. However, for the SSP1-2.6 scenario, decreases in CDD are combined with the increases in CWD and SDII over parts of ICP, indicating a generally wetter future in this region. The opposite patterns in CDD and CWD are projected over the Maritime Continent in this scenario, while the other four indices show prominent increases, especially over SUM and KAL. This nonuniform phenomenon suggests the intensification of both wet and dry conditions over SEA, which is consistent with the earlier projected changes in CMIP5 under the RCP 2.6 scenario (Sillmann et al 2013b). The pronounced increase of total precipitation of wet days (PRCPTOT) is projected over most of SEA under all SSP scenarios. Over some regions of the southern SUM, the projected changes show decreases under Reduced PRCPTOT is consistent with the increased CDD over southern SUM, while indices of extremes frequency (R20mm) and extremes intensity (Rx5day and SDII) show a significant and robust increase over the entire SEA, implying a potential risk of intensified precipitation extremes in natural ecosystems under the accelerated emission scenarios.

Projected changes in precipitation extremes
Compared with the SSP1-2.6 scenario, more pronounced increases in Rx1day, Rx5day, SDII, R20mm and R95pTOT are projected over all of SEA under the SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios. In addition, it is noteworthy that spatial patterns of CDD and CWD have gradually extended northward to the ICP from lower-to higher-emission scenarios. Larger high-confidence areas of increasing CDD (decreasing CWD) can be particularly observed under the SSP3-7.0 and SSP5-8.5 scenarios. This spatial evolution implies more heavy and extreme precipitation events (Rx5day, R20mm and SDII) are expected to occur with increased emissions, and extended dry spells (CDD) and shortened wet spells (CWD) could be mainly attributed to enhanced locally heavy rainfall over a short time scale.
Generally, precipitation extremes are becoming more intense over SEA which is supported by future projections from CORDEX climate model ensembles (Ge et al 2019b). This suggests that increases in extreme precipitation are closely related to enhanced tropical convective precipitation, further concentrating rainfall in a very short period over SEA. This would have critical impacts on local water resources, food security, agricultural production etc, especially for those developing countries with large coastal population densities.

Dominant roles of precipitation extremes in different scenarios
To identify the regional response to the different scenarios, we highlight here the CMIP6 projected percentage changes in precipitation extremes (averaged over the land area within SEA) in figure 4. In general, the CMIP6 models exhibit a distinctly large spread in different scenarios. However, it is notable that the ensemble medians of the index changes are all greater than 0 except CWD, implying increasing trends of precipitation extremes compared with the reference period of 1985-2014.
The most significant increases in R20mm and R95pTOT are projected with magnitudes of 22% and 102% for the SSP5-8.5 scenario (figures 4(a) and (h)), while the ensemble median changes of Rx5day and SDII show consistent increases, with magnitudes of 16% and 11%, respectively (figures 4(b) and (c)). The increasing magnitudes of Rx1day range from 7% (SSP1-2.6) to 21% (SSP5-8.5) and PRCPTOT shows a similar increase with the magnitude about 4% in SSP1-2.6, SSP2-4.5 and SSP3-7.0 scenarios, but  is 7% in SSP5-8.5. On the other hand, the projected changes in CDD and CWD are nearly 0 under SSP1-2.6 scenario. However, CDD is projected to increase by 3%, 10% and 8% under the scenarios of SSP2-4.5, SSP3-7.0 and SSP5-8.5, respectively, whereas the CWD is suggested to experience a pronounced decrease by −3% in SSP2-4.5 and by −7% in SSP3-7.0 and SSP5-8.5. In addition, we noticed that the non-uniform spatial patterns in CDD and CWD are projected under the SSP1-2.6 and SSP2-4.5 scenarios. To examine the climate responses to the lower GHG emission scenarios in more regional details, percentage changes of these two indices averaged over each subregion are presented in figure 5. The ensemble median changes in CDD indicate an increase over the PH, SUM and KAL, with magnitudes of 3%, 10% and 8% under the SSP1-2.6 scenario, while CWD shows a slight decrease over the SUM, KAL and SUL by −2%, −1% and −4%, respectively. For projections under the SSP-4.5 or higher emission scenario, CDD indicates consistent increases over the subregions, while it shows the reversed case for CWD. It implies that almost the entire area would be characterized by a moderate change under the lower GHG emissions, suggesting the importance of keeping the sustainability pathway for climate stabilization. By contrast, with the enhanced GHG pathways, the significant increase of R20mm and SDII suggest that wet days become wetter and intensified precipitation events more likely to occur frequently at the end of the 21st century, emphasizing that daily precipitation intensity tends to increase more abruptly than mean precipitation over the land area in SEA under a warmer future.

Conclusions and discussion
In this study, we have investigated future changes in precipitation extremes over SEA under four scenarios using current CMIP6 models. The results provide latest information on the climate responses of SEA to different emission scenarios at the end of the 21st century. The major findings are summarized as follows: On the other hand, the spatial evolution of CDD and CWD reveals that extended dry spells concurrent with shortened wet spells will gradually appear over SEA with rapidly rising emissions. (b) It is evident that the global mean temperature warming will increase the occurrence of precipitation extremes in the future. The precipitation over SEA is projected to be enhanced, while persistent increases would mainly concentrate on the tropical islands, suggesting the high sensitivity of the precipitation extremes to the GHG emissions. Most of the CMIP6 ensemble models agree with the sign of change, with SNRs all greater than 1, although they exhibit a relatively large spread in precipitation extremes. (c) The averaged median changes of indices vary considerably under different scenarios over SEA. The most pronounced increases are projected in R20mm and R95pTOT, with magnitudes of 22% and 102% under the SSP5-8.5 scenario, while the changes in CWD are projected to decrease, with the magnitude of −7% in SSP3-7.0 and SSP5-8.5 scenarios. The indices' evolutions indicate an intense increase in extreme precipitation events over this region in a warmer future.
The SEA climate is strongly affected by the tropical monsoon regime, and has been the subjected of increasing concern due to the high climate stresses of its exposure to global warming. Future projections in monsoon rainfall from CMIP3 to CMIP6 scenarios show its intensification over the Asian region, with a corresponding rise in precipitation extremes, which has many implications for ecological impacts and social risk throughout SEA (Scoccimarro et al 2013, Zhou et al 2014, Qi et al 2016, Zhang et al 2018, Chen et al 2020b, Grose et al 2020, Narsey et al 2020, Scoccimarro and Gualdi 2020. It is very disappointing that limiting warming to 1.5 • C seems to be barely feasible, although in fact the Paris Agreement target remains possible and could be attainable with ambitious and immediate action (Smith et al 2019). Based on our results from CMIP6, we again emphasize the necessity of restricting global warming for the mitigation of climate extremes in the countries in SEA. Further possible efforts should never be ignored and prompting actions would never be untimely.
In addition, any subsequent research should not only be conducted in terms of the frequency and intensity of climate extremes, but should also consider population exposure and local vulnerability (Chen andSun 2019, Chen et al 2020a). More work remains necessarily to be conducted on comparisons between model ensembles with and without bias correction (Maraun et al 2017, Guo et al 2018. High-resolution convection-permitting models and the new generation of COR-DEX experiment design for the dynamical downscaling of CMIP6 are fundamental for quantifying and assessing changes of climatological means, variability and extremes. In conclusion, some evidence has been shown in our present study using the currently released archive compared with the previous generation, which provides credible findings for regionally relevant climate change projections over SEA. However, we suggest that this result should be further examined as more models are added, since full CMIP6 models will be available in the near future and understanding the response to different global warming threshold is essential for eliminating the uncertainties in SEA climate projections.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https://esgf-node.llnl.gov/projects/cmip6/.

Acknowledgments
Two anonymous reviewers are thanked for their constructive comments that greatly improved the quality of the manuscript. We are grateful to Dr Haile Xue and Dr Caiyun Feng for inspiring discussions on the CMIP6 model ensemble. We also acknowledge SACA&D, ETCCDI, the German Climate Computing Center (DKRZ), and the climate modeling groups (listed in table 1) for their data and resources.